Overview
Amazon DynamoDB is used in this architecture as both the operational database for documents and metadata and the vector store for semantic retrieval.1 The unified design keeps structured records and their embeddings in one DynamoDB table rather than splitting them across a document database and a separate vector database.1 According to the article, this can reduce infrastructure cost and synchronization overhead while lowering the risk that retrieval uses stale vector data.1
Data and Search Design
The proposed table uses a composite primary key and stores embedding vectors as lists of numbers.1 It supports ordinary create, read, update, and delete operations as well as approximate-nearest-neighbor search for natural-language queries.1 The article describes using DynamoDB’s native vector-search capability and the SearchVectors API to retrieve semantically relevant content.1 A vector index can use a moderate-cardinality partition key, with category and tenant_id given as examples.1 Index setup requires waiting until the index is active, and early ValidationException responses may be retryable during that process.1
Role in an AI-Agent Architecture
In the example, an Amazon Bedrock agent routes requests to Lambda action groups, with one function performing semantic retrieval and another handling CRUD operations against the same table.1 A DynamoDB Streams-triggered embedding pipeline updates vectors when source content changes, keeping the searchable representation synchronized with the stored content.1 This pattern is aimed at knowledge-management systems that need both conversational search and direct document access.1
AI-agent uses
In a graduated-autonomy implementation, DynamoDB is used to persist agent trust state as part of the permission-governance architecture2.
FleetGPT example
Novus Hi-Tech’s FleetGPT uses Amazon DynamoDB for high-velocity fleet event streams, providing low-latency access alongside Amazon Aurora PostgreSQL for structured telemetry.23
Related Pages
- DynamoDB Vector Search — It describes how DynamoDB stores embeddings alongside operational records for semantic retrieval.
- FleetGPT — FleetGPT uses DynamoDB for low-latency access to fleet-event records.
- Amazon Bedrock — Amazon Bedrock and DynamoDB support complementary parts of FleetGPT: policy-document retrieval and fleet-event access.